SimplyFlo

@benedictk__ Would love to join in Munich!

Ok hear me out - wine and codex. We can set this up. Vienna, Munich, Zurich - who be keen? We (that is OpenAI) take over a wine bar and invite you (codex / coding power users) to talk codex and coding.

I didn’t know Grok could compile LaTeX too. The same goes for GPT-5.4 Pro. You can upload your .tex file, ask it to review it, and have it make a plan for how to amend it. I can provide you with the amended .tex file, internally checked to ensure that it compiles, including suggestions for compilation errors and even overfull or underfull \hbox warnings.

Grok has a nice feature: you can compile LaTeX directly in the chat. I asked it to draw the Torricelli point of a triangle in LaTeX, the point minimizing the sum of distances to the vertices, and to illustrate the proof with a picture.

... and 92m 45s.

Ok, it's getting ridiculous now, but again another record thinking time. 91m 44s. These times above 80 mins didn't show up until two weeks ago. And now they are the new normal? Is GPT-5.2 Pro getting smarter or just getting less computation?

If you're wondering what ChatGPT took so long: I'm working on this one right now. Or, more accurately: GPT-5.3 Pro is working on it, and I'm just assembling the LaTeX file: prism.openai.com/?u=b731c961-5c…

Still no 5.3 Pro, but a new "thought for" record on 5.2 Pro. 89m 37 s. Glade OpenAI has no token limit for Pro accounts.

And again a new record. Over 75m! Maybe the last one with GPT-5.2 Pro? GPT-5.3 Pro next?

GPT-5.2 Pro did what GPT-5.1 Pro couldn't - it set a new record and reasoned for more than 1 hour. In my case: 60m 4s. Great work, @OpenAI, no more cutoff at 1 hour.

@QuanticASI Math solution without god: n cancels out nnn = 100n + 10n + n = 111n. This means: 111n/3n=111/3=37

life has patterns that feel like GOD

And a new record: 74m

But since it works fine in Overleaf on the phone too, it might still be worth working on this one. What I still think is super important is full integration, especially of GPT-5.2 Pro with extended thinking. Basically my entire paper was written by switching between GPT-5 Pro / GPT-5.2 Pro and Overleaf. The way Prism is currently built, it doesn’t really work as a proper one-stop shop yet. But the concept is brilliant: the best parts of an online LaTeX editor combined with ChatGPT-level power. I believe the whole thing is setting the stage for upcoming pro models that will actually push math and physics forward. From OpenAI’s perspective it then makes total sense to offer everything science-related from a single source.

@kevinweil @vicapow On my macbook everything works perfectly, it compiles all the way through. So it looks like a bug that only happens on the phone because of the huge pdf. So it’s probably more of a rendering / memory issue on iOS Safari (WebKit).

💥 Today we’re introducing Prism—a free, AI-native workspace for scientists to write and collaborate on research, powered by GPT-5.2. Accelerating science requires progress on two fronts: 1. Frontier AI models that use scientific tools and can tackle the hardest problems 2. Integrating that AI into the products scientists use every day Prism is free to anyone with a ChatGPT account, with unlimited projects and collaborators. Try it today at prism.openai.com—would love to hear your feedback.

@kevinweil @vicapow What I did yesterday: Export a big tex-file that worked well in Overleaf as a standalone (bib integrated), import it to Prism. All with my mobile. Will try it again on my macbook in an hour or so. Will get back to you.

@kevinweil @vicapow Multiple errors while compiling. Pdf won't compile properly.

@SimplyFloJ Let us know specifically what you're seeing! cc @vicapow

Been having a blast with Claude Cowork for digitizing old family photos. The real pain point is EXIF metadata, especially dates. In my case, it’s a big family album spanning more than 60 years, and batch edits are basically useless for this kind of stuff. So I just gave Claude access to my scans and asked it to estimate dates based on people’s ages, plus the birthdates of the “usual suspects” who show up all the time. A few minutes later: surprisingly accurate timestamps on pretty much everything, and my whole library is now in proper chronological order. Doing that manually would’ve taken me hours. This took minutes.

@OpenAI GPT-5.2 Pro Extended Thinking did it again. New record, it thought for more than 66 minutes! 🤯

@mattshumer_ @OpenAI Yes, it is the first model that is able to reason for more than one hour! x.com/simplyfloj/sta…

Looks like @OpenAI has added an even MORE powerful version of Pro mode... you can now ask GPT-5.2 Pro to think even longer than before. Starting to test this... I have high expectations here.

@JonhernandezIA Yes, crazy. I got over 60m today.

WHAT THE ACTUAL F*CK?? maybe they went too far with pro? 😂

@koltregaskes ... and they dropped the 1h cutoff for reasoning. I just got over 1h reasoning time.

GPT-5. Pro now comes with two thinking models: Standard and Extended. Someone says the juice is 768 (xhigh)? I only see this on the web atm.